Now passing alpha/beta to the kernel as arguments as before fp16 support; in case of fp16 arguments are cast on host and in kernel

8 files changed, 10 insertions, 20 deletions

diff --git a/src/tuning/kernels/copy_fast.cpp b/src/tuning/kernels/copy_fast.cpp
index 34269bc7..78ded56e 100644
--- a/src/tuning/kernels/copy_fast.cpp
+++ b/src/tuning/kernels/copy_fast.cpp
@@ -86,11 +86,10 @@ class TuneCopy {
                            std::vector<T> &, std::vector<T> &,
                            std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentInput(a_mat);
     tuner.AddArgumentOutput(b_mat);
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
   }
 
   // Describes how to compute the performance metrics
diff --git a/src/tuning/kernels/copy_pad.cpp b/src/tuning/kernels/copy_pad.cpp
index 1e0dccd3..90f5ea82 100644
--- a/src/tuning/kernels/copy_pad.cpp
+++ b/src/tuning/kernels/copy_pad.cpp
@@ -86,7 +86,6 @@ class TunePad {
                            std::vector<T> &, std::vector<T> &,
                            std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(static_cast<int>(args.n));
     tuner.AddArgumentScalar(static_cast<int>(args.m));
@@ -97,7 +96,7 @@ class TunePad {
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(0);
     tuner.AddArgumentOutput(b_mat);
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
     tuner.AddArgumentScalar(0);
   }
 
diff --git a/src/tuning/kernels/transpose_fast.cpp b/src/tuning/kernels/transpose_fast.cpp
index 7ac19cb6..10fa80cb 100644
--- a/src/tuning/kernels/transpose_fast.cpp
+++ b/src/tuning/kernels/transpose_fast.cpp
@@ -91,11 +91,10 @@ class TuneTranspose {
                            std::vector<T> &, std::vector<T> &,
                            std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentInput(a_mat);
     tuner.AddArgumentOutput(b_mat);
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
   }
 
   // Describes how to compute the performance metrics
diff --git a/src/tuning/kernels/transpose_pad.cpp b/src/tuning/kernels/transpose_pad.cpp
index 63274415..507718eb 100644
--- a/src/tuning/kernels/transpose_pad.cpp
+++ b/src/tuning/kernels/transpose_pad.cpp
@@ -90,7 +90,6 @@ class TunePadTranspose {
                            std::vector<T> &, std::vector<T> &,
                            std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(static_cast<int>(args.n));
     tuner.AddArgumentScalar(static_cast<int>(args.m));
@@ -101,7 +100,7 @@ class TunePadTranspose {
     tuner.AddArgumentScalar(static_cast<int>(args.n));
     tuner.AddArgumentScalar(0);
     tuner.AddArgumentOutput(b_mat);
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
     tuner.AddArgumentScalar(0);
   }
 
diff --git a/src/tuning/kernels/xaxpy.cpp b/src/tuning/kernels/xaxpy.cpp
index 88d12c1f..0033b3c6 100644
--- a/src/tuning/kernels/xaxpy.cpp
+++ b/src/tuning/kernels/xaxpy.cpp
@@ -89,9 +89,8 @@ class TuneXaxpy {
                            std::vector<T> &x_vec, std::vector<T> &y_vec,
                            std::vector<T> &, std::vector<T> &, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.n));
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
     tuner.AddArgumentInput(x_vec);
     tuner.AddArgumentOutput(y_vec);
   }
diff --git a/src/tuning/kernels/xgemm.cpp b/src/tuning/kernels/xgemm.cpp
index 4b1efdef..898b8435 100644
--- a/src/tuning/kernels/xgemm.cpp
+++ b/src/tuning/kernels/xgemm.cpp
@@ -121,13 +121,11 @@ class TuneXgemm {
                            std::vector<T> &, std::vector<T> &,
                            std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &c_mat,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
-    auto beta_buffer = std::vector<T>{args.beta};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(static_cast<int>(args.n));
     tuner.AddArgumentScalar(static_cast<int>(args.k));
-    tuner.AddArgumentInput(alpha_buffer);
-    tuner.AddArgumentInput(beta_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
+    tuner.AddArgumentScalar(GetRealArg(args.beta));
     tuner.AddArgumentInput(a_mat);
     tuner.AddArgumentInput(b_mat);
     tuner.AddArgumentOutput(c_mat);
diff --git a/src/tuning/kernels/xgemv.cpp b/src/tuning/kernels/xgemv.cpp
index d42155ae..5c187d33 100644
--- a/src/tuning/kernels/xgemv.cpp
+++ b/src/tuning/kernels/xgemv.cpp
@@ -96,13 +96,11 @@ class TuneXgemv {
                            std::vector<T> &x_vec, std::vector<T> &y_vec,
                            std::vector<T> &a_mat, std::vector<T> &, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
-    auto beta_buffer = std::vector<T>{args.beta};
     auto a_rotated = (V==3) ? 1 : 0;
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(static_cast<int>(args.n));
-    tuner.AddArgumentInput(alpha_buffer);
-    tuner.AddArgumentInput(beta_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
+    tuner.AddArgumentScalar(GetRealArg(args.beta));
     tuner.AddArgumentScalar(static_cast<int>(a_rotated));
     tuner.AddArgumentInput(a_mat);
     tuner.AddArgumentScalar(0);
diff --git a/src/tuning/kernels/xger.cpp b/src/tuning/kernels/xger.cpp
index d2590c53..1fb5c531 100644
--- a/src/tuning/kernels/xger.cpp
+++ b/src/tuning/kernels/xger.cpp
@@ -85,10 +85,9 @@ class TuneXger {
                            std::vector<T> &x_vec, std::vector<T> &y_vec,
                            std::vector<T> &a_mat, std::vector<T> &, std::vector<T> &,
                            std::vector<T> &) {
-    auto alpha_buffer = std::vector<T>{args.alpha};
     tuner.AddArgumentScalar(static_cast<int>(args.m));
     tuner.AddArgumentScalar(static_cast<int>(args.n));
-    tuner.AddArgumentInput(alpha_buffer);
+    tuner.AddArgumentScalar(GetRealArg(args.alpha));
     tuner.AddArgumentInput(x_vec);
     tuner.AddArgumentScalar(0); // x_offset
     tuner.AddArgumentScalar(1); // x_increment